8 Lidars and wind turbine control

David Schlipf, Oliver Bischoff, Martin Hofsaß,Andreas Rettenmeier, Juan Jose Trujillo, and Martin KuhnEndowed Chair of Wind Energy, Institute of Aircraft Design,Universitat Stuttgart, Stuttgart, Germany

8.1 Introduction

Reducing mechanical loads caused by atmospheric turbulence and energy optimization in

the presence of varying wind are the key issue for wind turbine control. In terms of control

theory changes in the inflowing wind field as gusts, varying shears and directional changes

represent unknown disturbances. However, conventional feedback controllers can compensate

such excitations only with a delay since the disturbance has to be detected by its effects to

the turbine. This usually results in undesired loads and energy losses of wind turbines.

From the control theory point of view disturbance rejection can be improved by a feed-

forward control if the disturbance is known. Not fully covered by theory, but used in practice

is the further advantage of knowing the disturbance in the future, e.g. in chassis suspension

or in daily life when vision is used to circumnavigate obstacles with a bicycle.

In a similar way wind field measurements with remote sensing technologies such as lidar

might pave the way for predictive wind turbine control strategies aiming to increase energy

yield and reduce excessive loads on turbine components. Remote sensing offers wind speed

tracking at various points in space and time in advance of reaching the turbine and before

hitting sensors at the blades or nacelle. This provides the control and safety system with

sufficient reaction and processing time.

In Figure 75 the different steps for predictive wind turbine control are shown. The objective

of the first step is to obtain wind fields in different distances in front of the turbine, e.g. by

use of lidar. In the next step, turbulence theory, e.g. the “Taylor’s frozen turbulence theorem”,

is considered when modelling the wind on its way towards the turbine. In the last step, the

predicted future wind fields are used to improve wind turbine control by model predictive

control strategies.

wind

current

wind fields

in distance

future

wind fields

on rotor

enhanced

controlremote sensing

measurements

wind

modeling

predictive

control strategies

Figure 75: Steps in model predictive wind turbine control assisted by remote sensing

8.2 Measuring wind fields with lidar

Wind fields can be considered as time variant vector fields: one vector a in each point in space

p. Therefore the objective of wind fields assessment is to reconstruct wind fields in discrete

space and time points as much as possible.

Commercial lidar systems have the disadvantage that they are not flexible due to measure-

ment in conical domain and normally they are ground based and don’t measure in main wind

direction. Thus nacelle based lidar systems with scanner or beam splitter are better suited,

Risø-I-3068(EN) 137

but still some issues have to be considered:

• Which points have to be scanned to get best information for control purpose?

• How does the probe volume effect the measurements?

• How can 3D vectors be reconstructed from line-of-sight measurements to obtain infor-

mation for control? (see section 8.2)

To investigate these effects, a lidar simulator is presented in section 8.2.

The “Cyclops” dilemma

As a Cyclops cannot see three-dimensionally with only one eye, it isn’t possible to measure

a 3D wind vector with only one lidar system. Three lidar systems focusing in the same point

with linearly independent laser beams are needed. With one nacelle mounted lidar system,

the two missing systems can be substituted by different assumptions, e.g.:

1. no vertical and no horizontal wind component, or

2. no vertical component and homogenous flow on each height

In Figure 76 the effect of both assumption possibilities is shown. In this case the 3D vectors

in p1 and p2 (measured in the same height) should be reconstructed from the line-of-sight

wind speeds vlos1 and vlos2. The first assumption yield a11 and a21 representing a horizontal

shear. With the second assumption the resulting vectors a21 and a22 are equal representing

a cross-flow, as homogenous flow on each height was assumed.

The dilemma consist, if the lidar measurement should be used for yaw and pitch control

at the same time: If the first assumption is used to calculate the inhomogeneous inflow,

perfect alignment is assumed. If the second assumption is used to obtain the misalignment,

homogeneous flow is assumed.

vlos1vlos2

a11

a12a21

a22

p1 p2

Figure 76: Different possibilities of 3D wind vector reconstruction

WITLIS (WInd Turbine LIdar Simulator)

To understand better the above mentioned effects and to plan measurement campaigns it

is helpful to simulate lidar measurements. The main objective of the simulation tool is to

reproduce the operation of a nacelle-mounted lidar system. Thus the tool facilitates the

evaluation of scanning patterns to find the best hardware and software solution for applications

like control, power curve assessment and wake measurements. Wind turbine control strategies

based on lidar can be tested with aeroelastic wind turbine simulation tools in a realistic setup.

138 Risø-I-3068(EN)

A modular setup provides software parts that can be used to process measurement data in

the same way as simulated data (see Figure 77).

Figure 78 depicts the wind field reconstruction of a simulated measurement: On the left

side a generic wind field from TurbSim (Jonkman, 2009) is shown superposed with a wake for

better illustration. On the right side the interpolated wind field with WITLIS can be seen. The

dots represent a real trajectory as can be done by the adapted Windcube lidar and perfect

alignment is assumed. It can be seen that main characteristics can be measured.

wake model verification

with LIDAR

power curve

with LIDARturbine control

with LIDAR

WITLIS LIDAR

system

requirements

for wind fields

scanner recommendations

optimized scanning strategy

simulated / real

wind fields

scanning device restrictions

measurements

Figure 77: Interactions between the applications, the simulator and the lidar system

Figure 78: Simulation of realistic lidar measurements in a turbine wake: Original downflow

wind component (left) and scanned (right) downflow wind component

Risø-I-3068(EN) 139

8.3 Prediction of wind fields

The inflow wind field for control purpose shouldn’t be measured in the rotor plane because

of several reasons:

• Because of stability and performance reasons the undisturbed wind field should be mea-

sured. But wind in rotor plane is affected by the turbine itself.

• It is difficult to perform measurements in the rotor plane, e.g. with pitot tubes or lidar

on the blades.

• Measuring in front of the turbine provides the control system more time to react. The

easiest way to model the wind on its way towards the turbine is to use “Taylor’s frozen

turbulence theorem”. It assumes that wind characteristics remain the same while being

transported through space with the mean wind speed.

8.4 Improving control

The new information of upwind wind speeds obtained by lidar measurements can be used for

improving the turbines control systems. Thereby we distinguish between four different control

activities: Yaw control, speed control, collective pitch control and individual pitch control.

Possible benefit and potential is listed respectively in table 13.

Table 13: Possible application and benefit of lidar based control. Here is assumed that mea-

surements are at least as beneficial as (e)stimation.

benefit potential reference

yaw more energy up to 12% Cath the Wind (2009)

speed more energy up to 10%(e) Boukhezzar and Siguerdidjane (2005)

collective pitch less loads up to 20% Schlipf and Kuhn (2008)

individual pitch less loads up to 30%(e) Selvam et al. (2009)

In this paper we will focus on advanced collective pitch control only.

Predictive Disturbance Compensation (PDC)

Fluctuating wind speed causes the speed of rotation to vary, which affects the loads to the

turbine. The objective of pitch control in full operating load range is therefore to maintain a

constant rotational speed of the rotor. The wind speed data provided by a lidar can be used

to compensate wind speed fluctuations. The block diagram in figure 79 illustrates the control

schema.

−

vw

βΩrated Ω

eτs

ΣFB

ΣDCt→∞

ΣΩβ

ΣΩvw

Σ

Figure 79: Control loop with predictive disturbance compensation.

140 Risø-I-3068(EN)

The output y (in this case the rotor speed Ω) is influenced by two signals: pitch angle

β which is identically equal to the control input u and the effective wind speed veff which

is a disturbance d in terms of control theory. Subsequently the wind turbine system Σ can

theoretically be divided into two subsystems, Σyu and Σyd. If d is now forwarded by a transfer

function ΣDC to the control input u, a compensation of the disturbance can be reached. In

this case the feedback controller ΣFB is responsible for reference signal tracking and ΣDC for

disturbance rejection.

ΣDC = −Σ−1yuΣyd would give a perfect compensation, but in practice the inversion of the

nonlinear dynamic turbine model Σyu cannot be calculated. Therefore a static compensation

is proposed

ΣDCt→∞

= uss(dss)

which is the static value uss of the system input subject to the static disturbance dss. The

feedback controller ΣFB then will react during the dynamic transitions.

The time interval of the transition is influenced by the difference in the dynamic orders of

Σyd and Σyu. In the case of pitch control Σyu has a higher dynamic order, because pitch

angles have a delayed impact on the rotor speed compared to the wind disturbance. Therefore

a prediction time τ shifts the disturbance signal in time in the way that the pitch moves earlier.

Due to the lidar measurement in front of the rotor plane this prediction is possible.

Stability of the control loop is influenced neither by the added static feed-forward control

nor by time shift, because none of the newly implemented blocks is part of the closed control

loop and no new poles were introduced.

Simulation

For evaluation of the proposed controller a Simulink implementation of the generic aero-

elastic NREL offshore 5-MW baseline wind turbine model (Jonkman et al., 2009) was used

(see Table 14).

Table 14: Specification of the generic NREL wind turbine model.

Rotor upwind, 3 blades

Rated power output Prated = 5 MW @ Ωrated = 12.1 rpm

Dimensions D = 126 m, h = 90 m

Controller collective pitch with gain scheduling

Pitch actuator 2nd order

Filter Ω, β, vw : 1st order

For the PDC implementation the shift time τ is chosen to

τ = T63, filter wind + T63, pitch actuator = 1 s

where T63 denotes the rise time to 63 % of the filters and actuators final value respectively.

The static pitch over wind speed donated βss(vw,ss) is given in Jonkman et al. (2009) and

shown in Figure 80.

The wind is modeled using the stochastic, full-field, turbulent-wind simulator TurbSim

(Jonkman, 2009). The measuring of the full-field wind is simulated using WITLIS, see section

8.2. In order to use the wind data for collective pitch control, it is necessary to reduce it to

one effective wind speed veff. Therefore a weighting function (Figure 81) has been developed

Schlipf and Kuhn (2008). It takes account of the impact of the wind on the aerodynamic

torque with respect to the radius using Prandtl root and tip losses.

Risø-I-3068(EN) 141

β[deg]

vw [m/s]0 5 10 15 20 250

5

10

15

20

25

Figure 80: Static pitch over wind speed of the NREL turbine model.

δcp/δry[m]

z[m]

01

23

4 0.1

0.2

0.3

0.4

0.5

0.6

0.7

-50

0

50

-60-40-200204060

Figure 81: Weighting function for the calculation of the effective wind speed.

Results

The following simulation results show the impact of the proportional-integral (PI) controller

and the PDC on rotor speed and bending moment respectively. In frequency domain for PDC

a better disturbance rejection in the frequency range up to 0.3 Hz can be observed (figure 82),

in which according to the Kaimal spectrum the wind contains most of its energy.

Magnitude(dB)

PI

Static DC

Predictive DC

Frequency (Hz)10−3 10−2 10−1 100

-80

-60

-40

-20

0

Figure 82: Frequency domain: disturbance rejection of different control strategies.

142 Risø-I-3068(EN)

Extreme Operating Gust (EOG)

Figure 83 shows the results in time domain of simulations of an EOG according to IEC

(2005) applied on the NREL wind turbine model. Here, no turbulence occurs and perfect

measurement is assumed.

[m/s]/[]

[rpm]

time[s]

[MNm]

βPI

βPDC

vw

ΩPI

ΩPDC

Ωref

MPI

MPDC

0 5 10 15 20 25

0 5 10 15 20 25

0 5 10 15 20 25

-50

0

50

100

15011

12

13

14

5

10

15

20

Figure 83: Comparison of conventional and PDC control strategy (EOG). The subscript ref

denotes reference

PDC control technique leads to a significantly reduced rotor overspeed. Consequently the

fore-aft bending moment MyT at the tower base is decreased by PDC as well, compare

table 15.

Table 15: Standard deviation of signals from figure 83

PI PDC PDC/PI

σ(Ω) / [rpm] 0.74 0.08 9%

σ(MyT) / [MNm] 38.2 11.0 29%

Turbulent Wind Field

Figure 84 and table 16 show the results for simulations with realistic turbulent wind fields

with a spatial resolution of 9 m and a time resolution of 0.05 s. It is based on a Kaimal

spectrum with a mean wind speed at hub height of vH = 18 m s−1 and turbulence intensity

of TI = 16 %. Again, perfect measurement is assumed.

Table 16: Standard deviation of signals from figure 84

PI PDC PDC/PI

σ(Ω) / [rpm] 0.42 0.09 21%

σ(MyT) / [MNm] 12.8 8.72 68%

σ(β) / [deg/s] 0.60 0.47 78%

Risø-I-3068(EN) 143

[m/s]

time [s]

[rpm]

[]

vHH

vw

βPDC

βPI

ΩPDCΩratedΩPI

50 60 70 80 90 100 110 120 130 140 150

50 60 70 80 90 100 110 120 130 140 150

50 60 70 80 90 100 110 120 130 140 150

5

10

15

20

10

11

12

13

14

10

15

20

25

30

Figure 84: Comparison of conventional and PDC control strategy (turbulent wind field). vHH

is the hub height wind speed

PDC reduces rotor speed variation and loads at the tower base despite lower pitch dynamics

occur. Simulations also show reduced loads at the blades.

Measurement simulation using WITLIS

In a third simulation the influence of a non-perfect measurement, simulated by WITLIS (see

section 8.2) was investigated. From table 17 it can be seen that there is still a remarkable

improvement using PDC.

Table 17: Standard deviation of signals from simulation including WITLIS.

PI PDC PDC/PI

σ(Ω) / [rpm] 0.48 0.17 35%

σ(MyT) / [MNm] 15.1 11.8 87%

σ(β) / [deg/s] 0.79 0.65 83%

Robustness

A weaker performance of the PDC than the conventional PI controller is possible, if there are

errors which are not included in the simulation, e.g.

• inaccurate measurements of the wind speed

• wrong calculation of the effective wind speed

• incorrect static pitch curve

• errors in the model used

• invalidity of Taylor’s frozen turbulence theorem

• wrong estimate of the shift time τ

144 Risø-I-3068(EN)

Simulations with a varying parameter τ (figure 85) result in a wide range, where the

performance results of the PDC remain superior to the PI control.

τ [s]

σ(Ω

)[rpm]

PDCPI

-10 -8 -6 -4 -2 0 2 4 6 8 100

0.5

1

Figure 85: Standard deviation of Ω subject to different prediction time shifts τ for PDC and

PI control

Conclusions PDC

The proposed predictive disturbance compensation was presented as a new and powerful

control strategy for wind turbine control in full load range. PDC has a guaranteed stability

and implementation needs static pitch over wind speed information and one prediction time

parameter only.

The performed simulations indicate a significant decrease in rotor speed variation and tower

and blade loads without higher pitch actuator activity.

Further research concerns the other control objectives of table 13 as yaw control, speed

control and individual pitch control by use of lidar measurements.

Notationa horizontal shear of the flow

a wind vector

d disturbance

D rotor diameter

EOG extreme operating gust

h hub height

MyT tower fore-aft bending moment

p vector point in space

Prated rated power output

PDC predictive disturbance compensation

PI proportional-integral (controller)

TX X% rising time

TI turbulence intensity

u control input

veff effective wind speed

vHH hub height wind speed

vlos line-of-sight wind speed

WITLIS wind turbine lidar simulator

xSS steady state of variable x

y system output

β pitch angle

σX standard deviation of a variable X

Σ a model system or a subsystem, e.g. a wind turbine

ΣFB feedback controller

ΣDC disturbance compensation

τ prediction time of a signal

Ωrated rated rotor speed

Risø-I-3068(EN) 145

ReferencesBoukhezzar B. and Siguerdidjane H. (2005) Nonlinear control of variable speed wind turbines without wind

speed measurement. Proc. of the 44th IEEE Conf. on Decision and Control and the European Control

Conf., Seville

Catch the Wind (2009) Boosting power production. Catch the Wind, Inc. 1:15 pp

IEC (2005) IEC 61400-12-1 Wind turbines - Design requirements. Int. Electrotechnical Commission

Jonkman B. J. (2009) TurbSim User’s Guide. Technical Report NREL/TP-500-46198, 1.5, 85 pp

Jonkman B. J., Butterfield S., Musial W., and Scott G. (2009) Definition of a 5-MW reference wind turbine

for offshore system development. Technical Report NREL/TP-500-38060, 1, 75 pp

Schlipf D. and Kuhn M. (2008) Prospects of a collective pitch control by means of predictive disturbance

compensation assisted by wind speed measurements. DEWEC 1:4 pp

Selvam K., Kane S., van Wingerden J. W., van Engelen T., and Verhaegen M. (2009) Feedback-feedforward

individual pitch control for wind turbine load reduction. Int. J. Robust Nonlinear Control 19:72–91

146 Risø-I-3068(EN)